Professional Documents
Culture Documents
1
2
4
11
13
15
25
27
30
1. Introduction
Let M be a moduli space parametrizing objects P that come with C vector spaces
H i (P ) (i Z).
For example, some cohomology theory. The cohomology jump loci of M are defined
set-theoretically as
Vki = {P M | dim H i (P ) k}.
For every i, there is a chain of inclusions
M = V0i V1i V2i . . .
In many situations, the cohomology jump loci form closed subsets of the moduli
space. When the moduli is an algebraic variety or a scheme, the cohomology jump
loci are usually closed subschemes. Cohomology jump loci are at least as complex as
the moduli, and in general everything can be a moduli space. Hence the study of the
cohomology jump loci can be described as finding a balance between general setups one
can focus on and interesting properties one can draw.
1
In this article one finds a survey of the recent results related to cohomology jump
loci. It emphasizes connections with deformations with cohomology constraints, global
structural results for rank one local systems and line bundles, some connections with
restrictions on fundamental groups and homotopy types, and connections with classical
singularity theory and Bernstein-Sato ideals.
There are many results left out. For example, we do not survey the all the implications
on quasi-projective groups one can draw from the global structure of cohomology jump
loci of rank one local systems. We also do not survey the large body of work dedicated to
complements of hyperplane arrangements. One can find these two topics in the surveys
by Suciu [Su-2011, Su-2015].
Acknowledgement. The first author thanks Steve Zucker and the Department of
Mathematics of the Johns Hopkins University for the hospitality during the writing of
this survey. The first author was partly sponsored by NSA, FWO, a KU Leuven OT
grant, and a Flemish Methusalem grant.
2. Cohomology jump ideals
2.1. Old algebraic example: Fitting ideals. Let R be a commutative ring and M
a finitely generated R-module. Consider a free finite rank presentation of M
d
Rn Rm M 0.
Define the Fitting ideals of M
Jk (M) R
as the ideals
Jk (M) = Imk+1 (d)
generated by the minors of size m k + 1 of the matrix d. We set Jk (M) = 0 for k 0,
and Jk (M) = R for k > m. One has the classical result of Alexander and Fitting, see
[Ei-1995, 20.2]:
Theorem 2.1. Let R be a commutative ring and M a finitely generated R-module. The
Fitting ideal Jk (M) depends only on M and k, and not on the presentation of M.
Let M = Spec(R). The ideals Jk (M) define closed subschemes
Vk (M) M.
There is a chain of closed embeddings
M = V0 (M) V1 (M) . . . Vm+1 (M)
Thinking of M as the moduli space of its closed points, each closed point comes with
an assigned vector space
(m = maximal ideal of R) 7 M R R/m.
Moreover, Vk (M) is set-theoretically
Vk (M) = {m R maximal ideal | dimR/m (M R R/m) k},
hence it is a cohomology jump locus.
2.2. Cohomology jump ideals are generalized Fitting ideals. In [BW-2013], the
construction and the key properties of the Fitting ideals were generalized simultaneously
in two directions: to define invariants of a module from a whole free resolution, not just
from a free presentation, and to define these invariants for complexes of modules.
q
Let R be a noetherian commutative ring, and let M be a complex of R-modules,
q
bounded above. Suppose H i (M ) is a finitely generated R-module. There always exists
q
a bounded above complex F of finitely generated free R-modules and a morphism of
complexes
q
q
g : F M
q
which is a quasi-isomorphism. We define the cohomology jump ideals of M to be
Jki (M ) = Irank(F i )k+1 (di1 di )
q
di : F i F i+1
and
are the differentials of the complex F . The same conventions for ideals generated by
minors as in the previous subsection apply when rank(F i ) k + 1 falls outside the range
of the size of minors.
q
The cohomology jump loci of a bounded above complex M of R-modules with
finitely generated cohomology are the closed subschemes
q
Vki (M ) Spec(R)
q
defined by the cohomology jump ideals Jki (M ). In some sense, this is the mother of all
examples of cohomology jump loci.
q
and
Since taking determinantal ideals commutes with base change, one has the following
base change property of the cohomology jump ideals:
Corollary 2.5. Let R and M be defined as above, and let S be a noetherian R-algebra.
q
Moreover, suppose M is a complex of flat R-modules, then
q
Corollary 2.6. Suppose M is a complex of flat R-modules. Then for any maximal
q
ideal m of R, Jki (M ) m if and only if
q
dimR/m H i (M R R/m) k.
q
This says that the cohomology jump ideals deserve their name.
Remark 2.7. (Tor jumping loci) If M is a finitely generated R-module, the ideals Jki (M)
put a closed subscheme structure on the Tor jumping loci:
Vki (M)red = {P Spec(R) closed point | dimR/mP TorR
i (M, R/mP ) k}.
Remark 2.8. (Syzygies) If M is a finitely generated R-module and F is a free resq
olution, the syzygies of M are defined as the kernels of the differentials of F . Hence
the cohomology jump loci of a syzygy of M coincide with the far-enough-to-the-left
cohomology jump loci of M.
q
2.3. Dimension bounds. Recall the following classical homological algebra statement
from [Ei-1995, Theorem 20.9]:
Theorem 2.9. Let R be a ring. Let
d0
dn1
0 F 0 F 1 . . . F n1 F n
be a complex of free finitely generated R-modules. Then this complex is exact if and only
if
rank(F i ) = rank(di1 di )
and
depth Irank(di ) (di ) n i.
By applying this statement to the exact complex (F [1] F , d[1] d), one has
the following universal bound for cohomology support loci:
q
Theorem 2.10. With the same assumptions as in the previous theorem, if F is exact,
then
q
depth J1i (F ) n i.
q
3. Deformation theory
3.1. Local study. By a deformation problem we mean describing the formal scheme
or the analytic germ M(P ) at some object P of some moduli space M. A deformation
problem is equivalent to describing the corresponding functor, which we denote also by
M(P ) :
M(P ) : ART SET,
where ART is the category of Artinian local algebras of finite type over k, and SET is
the category of sets. Assuming M is a k-scheme, on objects the functor is defined as
A 7 Homksch (Spec(A), M(P )) = Homkalg(OM(P ) , A).
This functor is usually well-defined even if the moduli space is not. That is, by passing
to the categorical language, one can still ask about the deformations of some object
P inside a category. For example, one can ask about the deformations of a coherent
sheaf on an algebraic variety, regardless of any stability conditions necessary for the
construction of moduli spaces.
Suppose now the objects parametrized by M come with a cohomology theory, so that
one has the stratification of M by the cohomology jump loci Vki . The study of the formal
or analytic germs of the cohomology jump loci at an object P amounts to studying the
associated functor on Artinian local algebras as a subfunctor
i
Vk,(P
) M(P ) .
Even if the moduli space does not exist, the corresponding deformation problem with
cohomology constraints is usually well-defined.
3.2. Differential graded Lie algebras. In a letter written by P. Deligne to J. Millson
in 1986 in connection with the paper [GM-1988] by Goldman-Millson, a principle of
sweeping generality was stated. The letter is available on J. Millsons webpage [De-1986].
Delignes Principle. Every deformation problem over a field of characteristic zero
is controlled by a differential graded Lie algebra, with two quasi-isomorphic differential
graded Lie algebras giving the same deformation theory.
We explain next the meaning of Delignes Principle.
From now on, k = C. We use the abbreviation DGLA for a differential graded Lie
algebra.
Definition 3.1. A DGLA consists of the following set of data,
L
(1) a graded vector space C = iN C i over C,
(2) a Lie bracket which is bilinear, graded skew-commutative, and satisfies the graded
Jacobi identity, i.e., for any C i , C j and C k ,
[, ] + (1)ij [, ] = 0
and
(1)ki [, [, ]] + (1)ij [, [, ]] + (1)jk [, [, ]] = 0
(3) a family of linear maps, called the differential maps, di : C i C i+1 , satisfying
di+1 di = 0 and the Leibniz rule, i.e., for C i and C
d[, ] = [d, ] + (1)i [, d]
P
where d = di : C C.
We denote this DGLA by (C, d), or C when there is no risk of confusion. A homomorphism of DGLAs is a linear map which preserves the grading, Lie bracket, and the
differential maps.
Definition 3.2. A homomorphism of DGLAs g : C D is 1-equivalent if it induces
an isomorphism on cohomology up to degree 1 and a monomorphism at degree 2. Two
DGLAs are of the same 1-homotopy type, if they can be connected by a zig-zag of
1-equivalences.
The nicest DGLAs form a particular class:
Definition 3.3. For a DGLA (C, d), the cohomology forms a natural DGLA (H (C), 0)
with zero differential. We say the DGLA C is 1-formal if C is of the same 1-homotopy
q
type as (H (C), 0).
q
Given a DGLA (C, d), one can attach naturally a functor, called the deformation
functor,
Def(C, d) : ART SET
defined on Artinian local rings by
Maurer-Cartan elements of C C mA
A 7
,
gauge
that is,
{ C 1 C mA | d + 12 [, ] = 0}
.
A 7
C 0 C mA
Here mA is the maximal ideal of A, and C C mA is naturally a DGLA by letting
[ a, b] = [, ] ab
d( a) = d a.
Since (C C m)0 = C 0 C m is a nilpotent Lie algebra, the Campbell-Hausdorff multiplication defines a nilpotent Lie group structure on the space C 0 m. The gauge action
of an element C 0 m on C 1 m is given by
1 exp(ad )
7 exp(ad ) +
(d)
ad
in terms of power series.
Suppose now one has a deformation problem M(P ) over C as in 3.1. To say that a
DGLA C controls this deformation problem means that the two functors
M(P ) ) : ART SET
and
are isomorphic:
M(P ) = Def(C).
Constructing a DGLA responsible for a deformation problem is an ad-hoc process with
no all-encompassing recipe. Typically the DGLAs one first constructs in deformation
theory are infinite-dimensional, and there might be multiple choices of such constructions. The second statement of Delignes Principle is then crucial:
Theorem 3.4 ([GM-1988]). The deformation functor Def(C) only depends on the 1homotopy type of C. More precisely, if a morphism of DGLA f : C D is 1-equivalent,
then the induced transformation on functors f : Def(C) Def(D) is an isomorphism.
This finishes the explanation of Delignes Principle.
Example 3.5. ([FIM-2012]) For a complex manifold or complex algebraic variety X,
the infinitesimal deformations of an OX -coherent sheaf E are controlled by the DGLA
of global sections
q
q
q
(X, A (End (E )))
q
q
q
of any acyclic resolution A of the sheaf of DGLAs End (E ) of a locally free resolution
q
q
E of E. If X is a complex manifold, then A can chosen to be the Dolbeault resolution.
Remark 3.6. A theorem due to Lurie [Lu-2011] and Pridham [Pr-2011] in the framework
of derived algebraic geometry states that, with the appropriate axiomatization, Delignes
Principle is a theorem. However, in this axiomatization the notion of a deformation
problem might be different than the classical notion we have defined, and a classical
deformation problem might not come from any new type of deformation problem.
3.3. Deformations with constraints. By a deformation problem with cohomology constraints we mean the study of formal germs of cohomology jump loci
i
Vk,(P
) M(P )
for all i and k at once. This is equivalent to describing the corresponding functors on
i
local Artin rings which we denote also by Vk,(P
) , and which in fact are usually well-defined
even if the formal germs are not.
In [BW-2013], a generalization of Delignes Principle has been proposed and illustrated
in several examples:
New Principle. Every deformation problem with cohomology constraints over a field
of characteristic zero is controlled by a pair consisting of a DGLA together with a DGL
module, with two quasi-isomorphic pairs giving the same deformation theory with cohomology constraints.
Let us explain what this means. The deformation problem without constraints M(P )
comes, as claimed by Delignes Principle, with a DGLA C. Typically this is some kind
of endomorphism object. That is, C does not come alone, but it comes with a natural
module M on which it acts. Given a pair (C, M) consisting of a DGLA and DGL module,
[BW-2013] defines canonicallly cohomology jump deformation functors
Def ik (C, M) : ART SET.
Then
as subfunctors of M(P )
i
i
for all i, k
Vk,(P
) = Def k (C, M)
= Def(C). We recall now the basic details.
(H (C), H (M))
q
(M C A, d )
with
d := d idA +.
Moreover, (M C A, d ) has finitely generated cohomology over A.
In particular, one can apply the construction of cohomology jump ideals from 2.2 to
obtain ideals
Jki (M C A, d ) A.
The cohomology jump deformation functors of the DGLA pair (C, M)
Def ik (C, M) : ART SET
are defined by
{ C 1 C mA | d + 12 [, ] = 0 and Jki (M C A, d ) = 0}
.
C 0 C mA
This gives a well-defined subfunctor of Def(C), see [BW-2013]. Theorem 3.4 admits the
following refinement:
A 7
Theorem 3.13. The cohomology jump functor Def ik (C, M) only depends on the number
k and on the i-homotopy type of (C, M).
3.5. Finite-dimensional DGLA pairs. In the presence of finite dimensional DGLA
pairs, the cohomology jump deformation functors are pro-represented by schemes with
explicit equations, up to the gauge action. We explain what this means.
Let (C, M) be a finite dimensional DGLA pair. Fixing a base e1 , . . . , eb of C 1 , denote by x1 , . . . , xb the dual base. We identify the affine coordinate ring OC 1 with the
polynomial ring C[x1 , . . . , xb ]. The universal element of C 1 is defined as the element
wuniv =
b
X
ei xi
in C 1 C OC1 .
i=1
containing the origin. This is also called the versal deformation, or Kuranishi space of
C.
Def(C) is pro-represented by F (C)(0) modulo the gauge action when this quotient
exists in the category of schemes. This issue does not arise when, for example, the
gauge action of C 0 on C 1 is trivial. In that case,
Def(C)
= F (C)(0) .
If C is 1-formal, then the universal Maurer-Cartan equations become the quadratic
(that is, degree at most two) equations
[wuniv , wuniv ] = 0.
In this case, F (C) is a quadratic cone and we denote it by Q(C).
On F (C), one can define the OF (C) -linear universal Aomoto complex
(M C OF (C) , dM id + wuniv ).
q
10
11
with the natural DGLA structure. The deformations of E with cohomology constraints
are governed by the DGLA pair
0,
(A0,
Dol (End (E)), ADol (E)).
The main issue arising when applying the deformation thechniques of 3.1 - 3.4 is that
the DGLA pairs involved are usually infinite-dimensional and not formal, even in the
case of smooth projective curves. We will focus here on cases when formality holds.
Consider the moduli space M of stable rank n holomorphic vector bundles E with
vanishing Chern classes on a compact Kahler manifold X. We consider the Hodgetheoretic flavored cohomology jump loci
Vkpq (F ) M
given by
Theorem 4.1 ([D+1975], [Si-1992]). Let X be a compact Kahler manifold. For any
E M, the DGLA pair
p,
(A0,
Dol (End (E)), ADol (E F ))
is formal.
Let
Q(E) = { H 1 (X, End (E)) | = 0 H 2 (X, End (E))},
p
q
Rpq
k (E; F ) = { Q(E) | dim H (H (X, E F X ), ) k}.
The set Q(E) is endowed with a natural closed scheme structure defined using the
corresponding universal quadratic equations on the affine space H 1 (X, End(E)). In
particular, Q(E) has quadratic singularities. The set Rpq
k (E; F ) has a natural closed
subscheme structure of Q(E) defined using the cohomology jump ideals as in 3.5 of the
universal complex over Q(E) with OQ(E) -linear maps given by the cup-product:
(H (X, E F pX ) C OQ(E) , univ ).
q
12
Note that this is a complex of free finitely generated OQ(E) -modules such that the entries
of the matrices representing the maps in the complex come from linear polynomials in
C[H 1 (X, End (E))]. More precisely, Q(E) and Rpq
k (E; F ) are the quadratic cone and,
respectively, the resonance varieties as defined in 3.5 of the cohomology DGLA pair
with trivial differentials
(3)
This isomorphim means that one has explicit simple equations describing locally at
E the cohomology jump loci Vkpq (F ), via the universal equations mentioned above for
Rpq
k (E; F ). Note that although the equations are simple, the outcome is not necessarily
so.
The isomorphism
(M)(E)
= Q(E)(0) ,
which is a particular case of Theorem 4.2, was shown by Nadel [Na-1988] and GoldmanMillson [GM-1988]. For rank n = 1 bundles, the result is due to Green-Lazarsfeld
[GL-1987, GL-1991].
By fixing the attention on a generic vector bundle E laying in Vkpq (F ), that is, if
dim H q (X, E F pX ) = k,
then the generators of the cohomology jump ideals defining Rpq
k (E; F ) are precisely all
the entries of the matrices defining the cohomology jump ideals. Since these entries are
linear forms, one obtains:
Corollary 4.3. With the same assumptions as in Theorem 4.2, if
dim H q (X, E F pX ) = k,
then Vkpq (F ) has quadratic algebraic singularities at E.
This was also showed for F pX = OX by Martinengo [Mar-2009] and B. Wang
[Wa-2012].
Remark 4.4. The assumptions in the Theorem are made such that the controlling
DGLA pair is formal and the gauge action of H 0 is trivial on H 1 . As explained in
3.5, whenever this is the case, the quadratic cone and the resonance varieties of the
cohomology DGLA pair describe locally the moduli space and the cohomology jump
loci. We will present repetitions of this type of thinking to obtain similar results in
completely different contexts. This is how one puts the New Principle of 3.3 to work.
13
14
Here, for a local system L on X, the de Rham complex ADR (L) is the complex of global
L-valued C forms with the usual differential. Also, End (L ) is the local system of
endomorphisms of L .
To make this completely true, one needs to reintroduce in the picture the choice of
base point x. Let
q
: ADR (End (L )) g = End (L )|x
be the restriction map. This is a DGLA map, hence we can also define
q
as in 3.4, such that they describe the local structure of Vki (W ) at , [BW-2013].
Remark 5.2. In the eye-opening paper [DP-2014], Dimca and Papadima have addressed
the reduced local structure at the trivial representation of cohomology jump loci of
q
representations of fundamental groups. In this case, the de Rham complex ADR (X) of
X plays the controlling role, as it is a module over itself. The results in [BW-2013] were
the outcome of the desire of the authors to extended the results of [DP-2014] beyond
the reduced structure and beyond the trivial representation. The solution in terms of
DGLA pairs to infinitesimal deformation problems with cohomology constraints seemed
natural to us only after reading [DP-2014].
In cases where formality of the (augumented) DGLA pair is present, one can write
explicit local equations for Vki (W ) as explained in 3.5. Let
Q() = { Z 1 (1 (X), gl(n, C)ad ) | = 0 H 2 (X, End(L ))},
Rik (, W ) = { Q() | dim H i (H (X, L C W ), ) k},
with the natural scheme structures, where Z 1 stands for the vector space of 1-cocycles,
ad is the adjoint representation, and is the image of in cohomology.
Similarly, for a local system L on X, define
q
15
16
6.1. Local systems vs. Higgs line bundles. We recall the following from [Si-1992].
Let X be a compact Kahler manifold. Let
MDol (X) = {(E, ) | Higgs line bundle} = Pic (X) H 0 (X, 1X ),
where
Pic (X) = ker{c1 : Pic(X) H 2 (X, R)}.
The space MDol (X) is endowed with a natural complex structure. There is an isomorphism of real analytic groups
The cohomologies of rank one local systems and of Higgs line bundles are related via:
Theorem 6.1. ([GL-1987, Be-1988, GL-1991, Si-1992]) Let X be a compact Kahler
manifold. Let L MB (X) be a rank one local system with associated Higgs line bundle
(E, ) MDol (X). Then
M
H i (X, L)
H pq (E, ).
=
p+q=i
This has been extended by Simpson beyond the rank one case in [Si-1992, Si-1994].
We fix in this section the following notation about cohomology jump loci:
ik = {L MB (X) | dim H i (X, L) k}
Vkpq
Vki
Hkpq
pX )
MB (X),
k}
Pic (X),
Pic (X),
MDol (X).
17
Remark 6.2. Since Pic (X) = UDol (X) MDol (X), one has that
Vkpq = Hkpq Pic (X).
Arapuras trick [Ar-1992] shows, via the theorem above, that every irreducible component of a Hkpq comes from an irreducible component of ik .
Therefore we can take the point of view that studying line bundles with torsion first
Chern class is a particular case of studying rank one local systems.
6.2. Local study. Let X be a complex manifold. Consider the local study of the loci
ik MB (X) at a rank one representation . The DGLA pair describing the formal
schemes
ik,()
is
(ADR (X), ADR (L )),
q
Set-theoretically,
Rik = {w H 1 (X, C) | dim H i ((H (X, C), w .) k}.
q
This follows from making explicit for this case the section 3.5.
18
Remark 6.4. (Hyperplane arrangements I.) This result, for the reduced germs of ik at
the trivial representation, for the case when X is the complement of a hyperplane arrangement, had been obtained earlier by Esnault, Schechtman, and Viehweg [ESV-1992].
By Orlik-Solomon [OS-1980], the cohomology ring of X is a combinatorial invariant of
the hyperplane arrangement, that is, it only depends on the lattice of intersection of the
hyperplanes and not on their positions. Hence, the components through 1 of ik can be
detected combinatorially in this case from the resonance varieties Rik . In the context of
hyperplane arrangements, the resonance varieties have been introduced by Falk, see the
survey articles by Suciu referred to in the Introduction.
Remark 6.5. (Hyperplane arrangements II.) A very difficult folklore conjecture states
that all the cohomology jump loci ik of the complement of a hyperplane arrangement are
combinatorial invariants. A particular case of this conjecture states that the eigenvalues
of the monodromy on the Milnor fiber of a hyperplane arrangement are determined
combinatorially. The best result so far in this direction is the recent paper of Papadima
and Suciu [PaS-2014]. Surprisingly, they use the cohomology jump loci of rank two local
systems and resonance varieties defined over finite fields to address the rank one case.
Remark 6.6. The de Rham complex of X encodes the homotopy type of the topological
space X. A consequence is that the formal scheme
ik (X)(1)
at the trivial representation only depends on the i-homotopy type of X. For the underlying reduced germs, this is due to Dimca-Papadima [DP-2014]. The cohomology jump
deformation functors from 3.4 give the statement for the non-reduced structure as well.
Remark 6.7. (Gysin model) Typically ADR (X) is not formal. Hence the explicit deq
scription of ik (X)(1) depends on finding at least a finite model for ADR (X), that is,
q
a finite dimensional CDGA quasi-isomorphic with ADR (X). The crucial observation of
Dimca-Papadima [DP-2014] is that, if X is a smooth quasi-projective variety, there is
always such a model. This is the Gysin model of Morgan [Mo-1978], defined with respect
to any smooth projective compactification
q
j:X Y
S
with a simple normal crossings divisor Y \ X = iS Di as boundary. Here Di are the
q
irreducible components. For I S, let DI = iI Di . The Gysin CDGA A is defined
by setting
M
M
Ai =
Ap,q ,
Ap,q =
H p (DI , C).
p+q=i
|I|=q
19
The Gysin model A is in fact the (total complex of the) E2 page of the Leray spectral
sequence for the open embedding j:
q
20
Remark 6.13. The theorems build upon a long list of partial results. The chronology
is the following:
- Beauville [Be-1988] showed the linearity property of V11 for a compact Kahler manifold and conjectured the third theorem for Vkpq ;
- Green-Lazarsfeld [GL-1991] showed the linearity of Vkpq and 1k for a compact Kahler
manifold;
- Arapura [Ar-1992] showed that ik are finite unions of unitary translated subtori
for a compact Kahler manifold. He also showed that Hkpq are finite unions of unitary
translated subtori of Pic (X) times vector subspaces of H 0(X, 1X ), recovering thus the
result of [GL-1991].
- Simpson [Si-1993] proved the theorems for X smooth projective, namely by showing
that the components of ik , and hence of Vkpq , are torsion translated subtori;
- Arapura [Ar-1997] showed the first theorem. A small gap in the proof was filled later
in [AB+2013].
- Campana [Cam-2001] showed the third theorem for 1k ;
- Pink and Roessler [PR-2004] reproved the statement about Vkpq for a smooth projective variety via reduction modulo p;
- in [Bu-2009] we showed that ik UB (X), the cohomology jump loci of unitary rank
one local systems on a smooth quasi-projective variety X, and their Hodge theoretic
refinements, are finite unions of torsion translated subtori of Pic (X) times rational
convex polytopes. Combined with Theorem 6.10, this gives Theorem 6.11 under the
assumptions on the compactification as in Theorem 6.10, as remarked by Dimca and
Libgober.
- Dimca-Papadima [DP-2014] showed Theorem 6.9.
- The final statements of the last two theorems were after this just highly non-trivial
statements about existence of torsion points in each irreducible component of ik , this
being shown recently in [BW-2015] and, respectively, [Wa-2013].
The three theorems suggest that the following question might have a positive answer:
Question 6.14. Let X be the complement of a simple normal crossings divisor in a
compact Kahler manifold. Are the cohomology jump loci ik always finite unions of
torsion translated subtori?
6.4. Fibrations. The structure theorem is more refined in some cases. For a compact
Kahler manifold X, let
aX : X Alb(X)
be the Albanese map. For a compact analytic variety Y , we say that that Y has maximal
Albanese dimension if dim Y equals the dimension of the Albanese of any smooth model
of Y .
Theorem 6.15. ([GL-1991]) Let X be a compact Kahler manifold. Let Z be a positive
dimensional irreducible component of Vki Pic (X). Then there exists an analytic
dominant map
f :XY
with connected fibers to an analytic variety Y of maximal Albanese dimension and
dim Y i,
21
and with any smooth model Y of Albanese general type and with (Y ) > 0.
In the non-compact case, one has the following:
22
Let exdeg(X) be the exactness degree of this pair as defined in 3.6. Here the quadratic
cone of the pair is the affine space
Q = H 1 (X, OX ).
The universal Aomoto complex in this case is
(H (X, OX ) C OQ , wuniv ).
q
23
Theorem 6.24. ([GL-1987]) Let X be a compact Kahler manifold. Then for a general
line bunde L Pic (X),
H i (X, L) = 0
for
Theorem 6.22 has been generalized to Hodge pieces and to local systems by Popa and
Schnell:
Theorem 6.25. ([PSc-2013]) Let X be a smooth complex projective variety of dimension
n.
The cohomology support loci V1pq Pic (X) satisfy
codim V1pq |p + q n| (aX ),
The cohomology support loci i1 MB (X) satisfy
codim i1 2(|i n| (aX )).
The first part in the theorem implies the second part by the correspondence between
local systems and Higgs line bundles. Here (aX ) is the defect of semi-smallness of the
Albanese map,
(aX ) = max(2l dim X + dim Al ),
lN
where
Al = {y Alb(X) | dim f 1 (y) l}.
A weaker statement for abelian varieties was also proved by Kramer-Weissauer [KW-2015].
Remark 6.26. We like to think of Theorem 6.25 as, again, a special case of the codimension bound (2). In this case, the controlling DGLA pair for V pq locally at the trivial
line bundle is
q
q
(H (X, OX ), H (X, pX )),
according to (3). The quadratic cone of this pair is Q = H 1 (X, OX ) again. The universal
Aomoto complex of this pair is
(H (X, pX ) OQ , wuniv ).
q
24
These are called GV sheaves. This bound is by duality related to the bound (2) obtained
by classical homological algebra. The second step is to note that the derived direct image
of the trivial line bundle under the Albanese map of X decomposes into special sheaves
as above by the Decomposition Theorem for proper maps. The special sheaves are:
Theorem 6.27. ([PSc-2013]) Let A be a complex abelian variety. Let M be a mixed
Hodge module on A with underlying filtered D-module (M, F ). Then
codim V1i (grFk M) i
for all k and i.
6.6. Application to abelian varieties and compact complex tori. ([PPS-2015])
For a stronger statement and a generalization of the previous theorem to compact complex tori, see the recent article of Pareschi, Popa, and Schnell [PPS-2015]. The GV
sheaves, and their related cousins the M-regular sheaves, that is, coherent sheaves F
satisfying the stronger inequality
codim V1i (F ) > i,
have been used to prove the following numerical characterizations in terms of plurigenera:
Theorem 6.28.
([CH-1999]) A smooth complex projective variety X is birational to an abelian
variety if and only if P1 (X) = P2 (X) = 1 and q(X) = dim X .
([PPS-2015]) A compact Kahler manifold X is bimeromorphic to a compact complex torus if and only if dim H 1 (X, C) = 2 dim X and P1 (X) = P2 (X) = 1.
6.7. Constructible complexes on tori. Let K be a C-constructible complex on a
complex manifold. Consider the relative hypercohomology jump loci of rank one local
systems
ik (K) MB (X)
given by
q
ik (K) = {L MB (X) | dim Hi (X, K C L) k}.
Theorem 6.29. ([GaL-1996]) Let K be a perverse sheaf on the affine torus X = (C )n .
Then ik (K) is a finite unions of translates of subtori of MB (X) and
codim i1 (K) i.
We expect the translates to be by torsion elements.
As a corollary, one has the following generic vanishing, the perverse sheaf counterpart
of Theorem 6.24.
Theorem 6.30. Let X (C )n be a closed subvariety of dimension d such that the
shifted constant sheaf CX [d] is a perverse sheaf on (C )n (e.g. a locally complete intersection). Then
(1)d (X) 0.
Remark 6.31. This result had appeared already in [LS-1991] and was attributed to
Laumon in [GaL-1996]. The positivity of the signed Euler characteristic has been later
conjectured for any closed subvariety of (C )n in [HS-2014]. Counterexamples to this
conjecture were recently constructed in [BW-2014].
25
26
A remarkable result of Cohen-Schenck [CS-2013], conjectured by Suciu, is that the resonance varieties R11 , as in Remark 6.4, determine combinatorially the ranks of the Chen
groups of the fundamental group of a complement of hyperplane arrangements:
Theorem 7.3. ([CS-2013]) Let k (G) be the rank of the k-th Chen group of the fundamental group G of a complement U of a hyperplane arrangement. Let hk be the number
of k-dimensional irreducible components of the resonance variety
R11 (U) = {w H 1 (U, C) | H 1 (H (U, C), w .) 6= 0}.
q
Then
m+k2
.
k (G) = (k 1)
hm
k
m2
X
27
q
q
Kahler manifold Y has the same cohomology ring, i.e., H (Y, Z)
= H (X, Z) as graded
commutative rings, then Alb(Y ) is not an abelian variety, and hence Y is not projective.
Voisin also gave other constructions so she can encode the information of in the
q
R-coefficient cohomology ring H (X, R). Thus the optimal result of [Vo-2004] is the
following.
Theorem 7.4. [Vo-2004] There exists a compact Kahler manifold X which is not of the
real homotopy type of any smooth projective variety.
In [Wa-2013], Voisins theorem is reproved using cohomology jump loci. A compact
Kahler manifold X is constructed which is very similar to Voisins example. Now, the
information of the endomorphism is now encoded in the cohomology jump locus 21 (X)
of rank one local systems. Thus, by Voisins argument using rational Neron-Severi group,
we obtain the following theorem.
Theorem 7.5. [Wa-2013] There exists a compact Kahler manifold X with the following
property. Suppose Y is another compact Kahler manifold with H 1 (Y, C)
= H 1 (X, C), or
equivalently there is an isomorphism of germs MB (Y )(0)
= MB (X)(0) . If the isomorphism MB (Y )(1)
= 21 (X)(1) , then Y is
= MB (X)(1) induces an isomorphism 21 (Y )(1)
not projective.
As mentioned in Remark 6.6, the pairs of germs ik (X)(1) MB (X)(1) are determined
by the real i-homotopy type of X for all i and k. Thus we can slightly improve Voisins
result to the following.
Corollary 7.6. There exists a compact Kahler manifold X that is not of real 2-homotopy
type of any smooth projective variety.
8. Small ball complements and Bernstein-Sato ideals
We now let X = Cn and let
f :XC
be a holomorphic function. For a point x f 1 (0), let
Uf,x = Bx (X \ f 1 (0)),
where Bx is a small ball centered at x in X.
Conjecture 8.1. Let Uf,x be the complement of a hypersurface singularity germ as
above. Then the cohomology jump loci of rank one local systems ik MB (Uf,x ) are
finite unions of torsion translates of subtori.
Remark 8.2. It is known that the small ball complements Uf,x are 1-formal spaces, see
[DH-1988]. Therefore Theorem 6.3 applies, and hence the conjecture is true for those
irreducible components of 1k (Uf,x ) containing the trivial rank one local system, that is,
those components are subtori.
Remark 8.3. This conjecture is stated erroneously as a theorem in [Li-2009]. We have
also believed it to be proved in loc. cit. until the writing of [BW-2013] when we realized
that the rather difficult Theorem 6.11 should in principle be easier to prove. A red flag
about [Li-2009] was raised then. Scouting for the details, we remarked the following.
28
such that
(4)
In this case,
MB (Uf,x ) = (C )r ,
and a rank one local system L on Uf,x is identified with the r-tuple (C )r given by
the monodromies around each of the branches.
Lemma 8.5. Let f and x be as above, satisfying (4). Let
diag : C MB (Uf,x ) = (C )r
be the diagonal embedding 7 (, . . . , ). Then intersection of the cohomology support
loci of rank one local systems
[
i1 (Uf,x ) MB (Uf,x )
i
with the image of the diagonal embedding is the set of eigenvalues of the Milnor monodromy on the cohomology of Ff,x .
Remark 8.6. When (4) does not hold, one has a similar statement after adjusting the
definition of the diagonal embedding. The lemma follows immediately from [Bu-2015,
Prop. 3.31, Thm. 4]. Note that, as pointed out by Liu-Maxim [LM-2014], in all
the statements in [Bu-2015] where the uniform support Suppunif F CX of the Sabbah
specialization complex appears, the unif should be dropped to conform to what is proven
29
in [Bu-2015]. There are two other different proofs of this lemma known to us: one using
the Leray spectral sequence for the Milnor fibration, the other one using perverse sheaves
related to the open embedding of Uf,x in Bx .
A different approach to the Monodromy Theorem was given by Malgrange and Kashiwara. For a holomorphic function f : X C, one defines the Bernstein-Sato polynomial
bf (s) to be the monic generator of the ideal of polynomials b(s) C[s] satisfying
b(s)f s = P f s+1 ,
for some operator
P OX [1 , . . . , n , s],
where OX are the holomorphic functions on X, and i = /xi are the usual derivations
with respect to a choice of coordinates on X = Cn . A non-trivial result of Bernstein and
Sato guarantees the existence of a non-zero bf (s) for any f . If f is a polynomial, then
OX can be replaced with the polynomial ring in n variables.
One has the classical main result about Bernstein-Sato polynomials, letting Exp() =
exp(2i):
Theorem 8.7. Let f : Cn C be a holomorphic function. Then:
([Ka-1976])The roots of bf (s) are negative rational numbers.
([Mal-1983, Ka-1982])The set
{Exp() | is a root of bf (s)} C
is the set of all eigenvalues of the Milnor monodromy on the cohomologies of
Milnor fibers Ff,x of f at points x f 1 (0).
We have taken in [Bu-2015] the point of view that this result of Malgrange and Kashiwara hides a more general statement about rank one local systems and we tried to
uncover what this statement is. Our answer is in terms of the Bernstein-Sato ideals
BF,x C[s1 , . . . , sr ]
attached to the collection of functions
F = (f1 , . . . , fr ),
Qr
with, for simplicity, f = i=1 fi as in (4). The ideal BF,x is defined as the the ideal
generated by the polynomials b C[s1 , . . . , sr ] such that
b(s1 , . . . , sr )
r
Y
i=1
fisi
=P
r
Y
fisi +1
i=1
30
affine torus is cut out by the equations ti = 1 for those i with fi (y) 6= 0. By eliminating
these equations, we define the uniform cohomology support loci
i1 (Uf,y )unif (C )r ,
all sitting inside the same affine torus for all y close to x. Consider the zero locus
Z(BF,x ) Cr
of the Bernstein-Sato ideal.
Theorem 8.8. ([Bu-2015]) With the notation as above,
[
i1 (Uf,y )unif ,
Exp(Z(BF,x ))
iZ
yf 1 (0) close to x
[AB+2013]
[ADH-2015]
[Be-1988]
[Bu-2009]
[Bu-2015]
[BW-2013]
[BW-2014]
[BW-2015]
[Cam-2001]
[Cat-1991]
[CH-1999]
[CS-2013]
[D+1975]
D. Arapura, Higgs line bundles, Green-Lazarsfeld sets, and maps of Kahler manifolds to
curves. Bull. Amer. Math. Soc. 26 (1992), no. 2, 310-314. 17, 20, 21
D. Arapura, Geometry of cohomology support loci for local systems. I. J. Algebraic Geom.
6 (1997), no. 3, 563-597. 19, 20, 21, 22
D. Arapura, M. Nori Solvable fundamental groups of algebraic varieties and K?ahler manifolds. Compositio Math. 116 (1999), no. 2, 173-188. 25
E. Arbarello, M. Cornalba, P.A. Griffiths, J. Harris, Geometry of algebraic curves. Vol. I.
Grundlehren der Mathematischen Wissenschaften 267. Springer-Verlag, New York, 1985.
xvi+386 pp. 11
E. Artal Bartolo, J.I. Cogolludo-Agustn, D. Matei, Characteristic varieties of quasiprojective manifolds and orbifolds. Geom. Topol. 17 (2013), no. 1, 273-309. 20, 22
D. Arapura, A. Dimca, R. Hain, On the fundamental groups of normal varieties.
arXiv:1412.1483. 21
A. Beauville, Annulation du H 1 et systemes paracanoniques sur les surfaces. J. Reine
Angew. Math. 388 (1988), 149-157. 16, 20, 21
N. Budur, Unitary local systems, multiplier ideals, and polynomial periodicity of Hodge
numbers. Adv. Math. 221 (2009), 217-250. 20
N. Budur, Bernstein-Sato ideals and local systems. Ann. Inst. Fourier 64 (2015). 28, 29, 30
N. Budur and B. Wang, Cohomology jump loci of differential graded Lie algebras.
arXiv:1309.2264. To appear in Compositio Math. 3, 7, 9, 12, 13, 14, 15, 27
N. Budur and B. Wang, The signed Euler characteristic of very affine varieties.
arXiv:1403.4371. To appear in Int. Math. Res. Notices. 24
N. Budur and B. Wang, Cohomology jump loci of quasi-projective varieties. Ann. Sci.
Ecole
Norm. Sup. 48 (2015), no. 1. 19, 20
F. Campana, Ensembles de Green-Lazarsfeld et quotients resolubles des groupes de Kahler.
J. Algebraic Geom. 10 (2001), no. 4, 599-622. 20, 25
F. Catanese, Moduli and classification of irregular Kaehler manifolds (and algebraic varieties) with Albanese general type fibrations, Inv. Math. 104 (1991), 263-289. 21
J. Chen and C. Hacon, Characterization of abelian varieties. Invent. Math. 143 (2001), no.
2, 435-447. 24
D.Cohen, H. Schenck, Chen ranks and resonance. arXiv:1312.3652. 26
P. Deligne, P. Griffiths, J. Morgan, D. Sullivan, Real homotopy theory of Kahler manifolds.
Invent. Math. 29 (1975), no. 3, 245-274. 11
[De-1986]
[Di-2007a]
[Di-2007b]
[DP-2014]
[DPS-2008]
[DPS-2009]
[DH-1988]
[Ei-1995]
[ESV-1992]
[FIM-2012]
[GaL-1996]
[GL-1987]
[GL-1991]
[GM-1988]
[HS-2014]
[Ka-1976]
[Ka-1982]
[KW-2015]
[Li-2009]
[Lu-2011]
[LM-2014]
[LP-2010]
[LS-1991]
31
P.
Deligne,
Letter
to
J.
Millson,
April
24,
1986,
http://www.math.umd.edu/~millson/papers/deligne.pdf 5
A. Dimca, Characteristic varieties and constructible sheaves. Rend. Lincei Mat. Appl. 18
(2007), 365-389. 22
A. Dimca, On the irreducible components of characteristic varieties, An. S
tiint. Univ.
Ovidius Constanta Ser. Mat. 15 (2007) 67-73. 22
A. Dimca, S. Papadima, Nonabelian cohomology jump loci from an analytic viewpoint.
Commun. Contemp. Math. 16 (2014), no. 4, 1350025, 47 pp. 13, 14, 18, 19, 20
A. Dimca, S. Papadima, A.I. Suciu, Alexander polynomials: essential variables and multiplicities, Int. Math. Res. Not. 2008. 22
A. Dimca, S. Papadima, A.I. Suciu, Topology and geometry of cohomology jump loci.
Duke Math. J. 148 (2009), no. 3, 405-457. 17
A. Durfee, R. Hain, Mixed Hodge structures on the homotopy of links. Math. Ann. 280
(1988), no. 1, 69-83. 27
D. Eisenbud, Commutative algebra. With a view toward algebraic geometry. Graduate
Texts in Mathematics, 150. Springer-Verlag, New York, 1995. 2, 4
H. Esnault, V. Schechtman, E. Viehweg, Cohomology of local systems of the complements
of hyperplanes. Invent. Math. 109 (1992), no. 3, 557-561; Erratum, ibid. 112 (1993), no. 2,
447. 18
D. Fiorenza, D. Iacono, E. Martinengo, Differential graded Lie algebras controlling infinitesimal deformations of coherent sheaves. J. Eur. Math. Soc. (JEMS) 14 (2012), no. 2,
521-540. 6
O. Gabber and F. Loeser, Faisceaux pervers l-adiques sur un tore. Duke Math. J. 83 (1996),
no. 3, 501-606. 24
M. Green, R. Lazarsfeld, Deformation theory, generic vanishing theorems, and some conjectures of Enriques, Catanese and Beauville. Invent. Math. 90 (1987), no. 2, 389-407. 12,
16, 21, 22, 23
M. Green, R. Lazarsfeld, Higher obstructions to deforming cohomology groups of line
bundles. J. Amer. Math. Soc. 4 (1991), no. 1, 87-103. 12, 16, 20, 21, 22
W. Goldman, J. Millson, Deformations of flat bundles over Kahler manifolds. Inst. Hautes
Etudes
Sci. Publ. Math. No. 67 (1988), 43-96. 5, 6, 12, 14
J. Huh, B. Sturmfels, Likelihood Geometry. Combinatorial Algebraic Geometry, Lecture
Notes in Mathematics 2108, Springer, 2014. 24
M. Kashiwara, B-functions and holonomic systems. Rationality of roots of B-functions.
Invent. Math. 38 (1976/77), no. 1, 33-53. 29
M. Kashiwara, Vanishing cycle sheaves and holonomic systems of differential equations.
Algebraic geometry (Tokyo/Kyoto, 1982), 134-142, Lecture Notes in Math., 1016, Springer,
Berlin, 1983. 29
T. Kramer, R. Weissauer, Vanishing theorems for constructible sheaves on abelian varieties.
J. Alg. Geom. 24 (2015) 531-568. 23
A. Libgober, Non-vanishing loci of Hodge numbers of local systems, Manuscripta Math.
128 (2009), no. 1, 1-31. 27, 28
J. Lurie, Formal Moduli Problems. 2011. http://www.math.harvard.edu/~lurie/papers/DAG-X.pdf
7
Y. Liu, L. Maxim, Characteristic varieties of hypersurface complements. arXiv:1411.7360.
28
R. Lazarsfeld, M. Popa, Derivative complex, BGG correspondence, and numerical inequalities for compact K
ahler manifolds. Invent. Math. 182 (2010), no. 3, 605-633. 22
F. Loeser, C. Sabbah, Caracterisation des D-modules hypergeometriques irreductibles sur
le tore. C. R. Acad. Sci. Paris Ser. I Math. 312 (1991), no. 10, 735-738. Caracterisation
des D-modules hypergeometriques irreductibles sur le tore. II. C. R. Acad. Sci. Paris Ser.
I Math. 315 (1992), no. 12, 1263-1264. 24
32
[Mal-1983]
[Mar-2009]
[Mar-2012]
[Mo-1978]
[Na-1988]
[OS-1980]
[PaS-2014]
[Pr-2011]
[PP-2009]
[PPS-2015]
[PR-2004]
[PS-2008]
[PSc-2013]
[Si-1992]
[Si-1993]
[Si-1994]
[Su-2011]
[Su-2015]
[Ti-1987]
[Vo-2004]
[Wa-2012]
[Wa-2013]
B. Malgrange, Polyn
omes de Bernstein-Sato et cohomologie evanescente. Analysis and
topology on singular spaces, II, III (Luminy, 1981), 243-267, Asterisque, 101-102, Soc.
Math. France, Paris, 1983. 29
E. Martinengo, Local structure of Brill-Noether strata in the moduli space of flat stable
bundles. Rend. Semin. Mat. Univ. Padova 121 (2009), 259-280. 12
E. Martinengo, Infinitesimal deformations of Hitchin pairs and Hitchin map. Internat. J.
Math. 23 (2012), no. 7, 125-153.
J.W. Morgan, The algebraic topology of smooth algebraic varieties. Inst. Hautes Etudes
Sci. Publ. Math. No. 48 (1978), 137204. Correction to: The algebraic topology of smooth
C. Simpson, Subspaces of moduli spaces of rank one local systems. Ann. Sci. Ecole
Norm.
Sup. (4) 26 (1993), no. 3, 361-401. 20
C. Simpson, Moduli of representations of the fundamental group of a smooth variety. II.
33